---
title: "Using Claude for legal work: what to use it for and what to check"
url: https://notionforlawyers.com/ai/claude
updated: 2026-10-17
author: Raghav Handa
---

# Using Claude for legal work: what to use it for and what to check

Claude, Anthropic's AI assistant, is often used by lawyers for reading and summarising long text, redrafting and structuring arguments. As with any assistant, output needs verification and client data should only be entered once your firm has reviewed the specific plan's current terms. Treat it as a drafting aid, not a source of authority.

## Key takeaways

- Good fits: summarising long non-confidential documents, redrafting, outlines and critique of your own drafts.
- Bad fits: unverified legal research, citations and anything requiring current law without checking.
- Data handling differs by plan; read Anthropic's current terms and privacy documentation for the plan you use.
- Review the output yourself. Responsibility stays with the lawyer.

## Where Claude fits in legal work

General information, not legal advice. Check your own bar or regulator's guidance and your client engagement terms before putting any client information into an AI tool.

Claude is a family of AI models from Anthropic, available through a chat app, an API, and tools such as Claude Code. Many users like it for working through long passages of text and for a measured writing style, but quality varies by model version and task, so test it yourself on anonymised material rather than relying on reputation.

- Summarise a long public report or article for a briefing note
- Critique your own draft: ask it to find ambiguity and missing definitions
- Turn rough notes into a structured memo outline
- Produce a client-friendly explanation of a concept
- Draft a checklist for a recurring process


## Data-control questions

Do not rely on this page for current terms. Ask these questions of Anthropic's published documentation for the plan you or your firm will use:

| Topic | What to confirm |
| --- | --- |
| Training use | Whether conversations may be used to improve models on your plan, and how to opt out |
| Retention | How long conversations and uploaded files are kept, and how deletion works |
| Business terms | Whether commercial or enterprise terms with confidentiality commitments are available |
| Access | Who at the vendor may review conversations and under what circumstances |
| Sharing | Whether shared chats or projects can be seen by others |


## Prompts that work well

Claude responds well to clear structure. Give it the role, the audience and the format, paste the anonymised text, and tell it to list its assumptions. Ask it to quote the passage it relies on so you can check it against the source.

Check the quotes: When an assistant quotes a document back to you, compare the quote with the original. Summaries can drift from the text.

- AI prompt library: https://notionforlawyers.com/ai/prompts
- Claude chat guide: https://notionforlawyers.com/library/guide-claude-chat
- Claude Code for lawyers: https://notionforlawyers.com/library/guide-claude-code


## Limits

Claude can make factual and legal errors with confidence, including invented citations. It does not know your jurisdiction's current rules unless you provide them, and even then it can misapply them. It is not a substitute for primary-source research or for professional judgment, and the duty of competence and confidentiality sits with you.


## FAQ

### Is Claude safer than ChatGPT for legal work?

We cannot claim that. Safety depends on plan terms, settings and your use. Compare each vendor's current documentation against your firm's requirements.

### Can Claude read an entire contract?

It can handle long text within its limits, but you should still verify its reading against the document, especially definitions and exceptions.

### Should I upload client files to Claude?

Only if your firm has approved the specific plan and the client's terms allow it. Otherwise anonymise or avoid it.

### What about Claude Code?

Claude Code is a coding agent for the command line. See our guide for cautious non-client-data uses.
